人脸生成（Face Generation）¶
在该项目中，你将使用生成式对抗网络（Generative Adversarial Nets）来生成新的人脸图像。
获取数据¶
该项目将使用以下数据集：
- MNIST
- CelebA
由于 CelebA 数据集比较复杂，而且这是你第一次使用 GANs。我们想让你先在 MNIST 数据集上测试你的 GANs 模型，以让你更快的评估所建立模型的性能。
如果你在使用 FloydHub, 请将 data_dir 设置为 "/input" 并使用 FloydHub data ID "R5KrjnANiKVhLWAkpXhNBe".
In [2]:
data_dir = './data'
# FloydHub - Use with data ID "R5KrjnANiKVhLWAkpXhNBe"
#data_dir = '/input'
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import helper
helper.download_extract('mnist', data_dir)
helper.download_extract('celeba', data_dir)
In [3]:
show_n_images = 25
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
%matplotlib inline
import os
from glob import glob
from matplotlib import pyplot
mnist_images = helper.get_batch(glob(os.path.join(data_dir, 'mnist/*.jpg'))[:show_n_images], 28, 28, 'L')
pyplot.imshow(helper.images_square_grid(mnist_images, 'L'), cmap='gray')
Out[3]:
CelebA¶
CelebFaces Attributes Dataset (CelebA) 是一个包含 20 多万张名人图片及相关图片说明的数据集。你将用此数据集生成人脸，不会用不到相关说明。你可以更改 show_n_images 探索此数据集。
In [4]:
show_n_images = 25
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
mnist_images = helper.get_batch(glob(os.path.join(data_dir, 'img_align_celeba/*.jpg'))[:show_n_images], 28, 28, 'RGB')
pyplot.imshow(helper.images_square_grid(mnist_images, 'RGB'))
Out[4]:
预处理数据（Preprocess the Data）¶
由于该项目的重点是建立 GANs 模型，我们将为你预处理数据。
经过数据预处理，MNIST 和 CelebA 数据集的值在 28×28 维度图像的 [-0.5, 0.5] 范围内。CelebA 数据集中的图像裁剪了非脸部的图像部分，然后调整到 28x28 维度。
MNIST 数据集中的图像是单通道的黑白图像，CelebA 数据集中的图像是 三通道的 RGB 彩色图像。
建立神经网络（Build the Neural Network）¶
你将通过部署以下函数来建立 GANs 的主要组成部分:
model_inputsdiscriminatorgeneratormodel_lossmodel_opttrain
检查 TensorFlow 版本并获取 GPU 型号¶
检查你是否使用正确的 TensorFlow 版本，并获取 GPU 型号
In [5]:
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
from distutils.version import LooseVersion
import warnings
import tensorflow as tf
# Check TensorFlow Version
assert LooseVersion(tf.__version__) >= LooseVersion('1.0'), 'Please use TensorFlow version 1.0 or newer. You are using {}'.format(tf.__version__)
print('TensorFlow Version: {}'.format(tf.__version__))
# Check for a GPU
if not tf.test.gpu_device_name():
warnings.warn('No GPU found. Please use a GPU to train your neural network.')
else:
print('Default GPU Device: {}'.format(tf.test.gpu_device_name()))
输入（Input）¶
部署 model_inputs 函数以创建用于神经网络的 占位符 (TF Placeholders)。请创建以下占位符：
- 输入图像占位符: 使用
image_width，image_height和image_channels设置为 rank 4。 - 输入 Z 占位符: 设置为 rank 2，并命名为
z_dim。 - 学习速率占位符: 设置为 rank 0。
返回占位符元组的形状为 (tensor of real input images, tensor of z data, learning rate)。
In [6]:
import problem_unittests as tests
def model_inputs(image_width, image_height, image_channels, z_dim):
"""
Create the model inputs
:param image_width: The input image width
:param image_height: The input image height
:param image_channels: The number of image channels
:param z_dim: The dimension of Z
:return: Tuple of (tensor of real input images, tensor of z data, learning rate)
"""
# TODO: Implement Function
inputs_real = tf.placeholder(tf.float32, (None, image_width, image_height, image_channels), name = 'input_real')
inputs_z = tf.placeholder(tf.float32, (None, z_dim), name = 'input_z')
learning_rate = tf.placeholder(tf.float32, name = 'learning_rate')
return inputs_real, inputs_z, learning_rate
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_inputs(model_inputs)
辨别器（Discriminator）¶
部署 discriminator 函数创建辨别器神经网络以辨别 images。该函数应能够重复使用神经网络中的各种变量。 在 tf.variable_scope 中使用 "discriminator" 的变量空间名来重复使用该函数中的变量。
该函数应返回形如 (tensor output of the discriminator, tensor logits of the discriminator) 的元组。
In [7]:
def discriminator(images, reuse=False):
"""
Create the discriminator network
:param image: Tensor of input image(s)
:param reuse: Boolean if the weights should be reused
:return: Tuple of (tensor output of the discriminator, tensor logits of the discriminator)
"""
# TODO: Implement Function
# Set up constant parameters
alpha = 0.2
keep_prob = 0.5
padding = 'SAME'
strides = 2 # Use same strides in the next three layers
# Apply convolution
with tf.variable_scope('discriminator', reuse = reuse):
# Conv layer 1
cl_1 = tf.layers.conv2d(images, 64, 4, strides, padding, kernel_initializer=tf.contrib.layers.xavier_initializer(uniform=True, seed=None, dtype=tf.float32))
# bn_1 = tf.layers.batch_normalization(cl_1, training=True)
lrelu_1 = tf.maximum(alpha * cl_1, cl_1)
dp_1 = tf.nn.dropout(lrelu_1, keep_prob)
# Conv layer 2
cl_2 = tf.layers.conv2d(dp_1, 128, 4, strides, padding, kernel_initializer = tf.contrib.layers.xavier_initializer(uniform=True, seed=None, dtype=tf.float32))
bn_2 = tf.layers.batch_normalization(cl_2, training = True)
lrelu_2 = tf.maximum(alpha * bn_2, bn_2)
dp_2 = tf.nn.dropout(lrelu_2, keep_prob)
# Conv layer 3
cl_3 = tf.layers.conv2d(dp_2, 256, 4, strides, padding, kernel_initializer = tf.contrib.layers.xavier_initializer(uniform=True, seed=None, dtype=tf.float32))
bn_3 = tf.layers.batch_normalization(cl_3, training = True)
lrelu_3 = tf.maximum(alpha * bn_3, bn_3)
dp_3 = tf.nn.dropout(lrelu_3, keep_prob)
# Conv layer 4
# cl_4 = tf.layers.conv2d(lrelu_3, 512, 4, strides, padding) # kernel_initializer=tf.contrib.layers.xavier_initializer(uniform=True, seed=None, dtype=tf.float32
# bn_4 = tf.layers.batch_normalization(cl_4, training=True)
# lrelu_4 = tf.maximum(alpha * bn_4, bn_4)
# dp_4 = tf.nn.dropout(lrelu_4, keep_prob)
# Flatten
flatten = tf.reshape(dp_3, (-1, 2*2*512))
# Logits
logits = tf.layers.dense(flatten, 1)
# Output value
output = tf.sigmoid(logits)
return output, logits
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_discriminator(discriminator, tf)
生成器（Generator）¶
部署 generator 函数以使用 z 生成图像。该函数应能够重复使用神经网络中的各种变量。
在 tf.variable_scope 中使用 "generator" 的变量空间名来重复使用该函数中的变量。
该函数应返回所生成的 28 x 28 x out_channel_dim 维度图像。
In [10]:
def generator(z, out_channel_dim, is_train=True):
"""
Create the generator network
:param z: Input z
:param out_channel_dim: The number of channels in the output image
:param is_train: Boolean if generator is being used for training
:return: The tensor output of the generator
"""
# TODO: Implement Function
# Set up constant parameters
reuse = not is_train
alpha = 0.2
keep_prob = 0.5
strides_1 = 1
strides_2 = 2
padding_1 = 'SAME'
padding_2 = 'VALID'
# Apply convolution
with tf.variable_scope('generator', reuse = reuse):
fc_1 = tf.layers.dense(z, 4*4*512)
# Fully connected layer 1
fc_1 = tf.reshape(fc_1, (-1, 4, 4, 512))
# bn_1 = tf.layers.batch_normalization(fc_1, training = is_train)
lrelu_1 = tf.maximum(alpha * fc_1, fc_1)
dp_1 = tf.nn.dropout(lrelu_1, keep_prob)
# Fully connected layer 2
fc_2 = tf.layers.conv2d_transpose(dp_1, 256, 4, strides_1, padding_2, kernel_initializer=tf.contrib.layers.xavier_initializer(uniform=True, seed=None, dtype=tf.float32))
bn_2 = tf.layers.batch_normalization(fc_2, training = is_train)
lrelu_2 = tf.maximum(alpha * bn_2, bn_2)
dp_2 = tf.nn.dropout(lrelu_2, keep_prob)
# Fully connected layer 3
fc_3 = tf.layers.conv2d_transpose(lrelu_2, 128, 4, strides_2, padding_1, kernel_initializer=tf.contrib.layers.xavier_initializer(uniform=True, seed=None, dtype=tf.float32))
bn_3 = tf.layers.batch_normalization(fc_3, training = is_train)
lrelu_3 = tf.maximum(alpha * bn_3, bn_3)
dp_3 = tf.nn.dropout(lrelu_3, keep_prob)
# Fully connected layer 4
# fc_4 = tf.layers.conv2d_transpose(dp_3, 32, 4, strides_2, padding_1)
# bn_4 = tf.layers.batch_normalization(fc_4, training=is_train)
# lrelu_4 = tf.maximum(alpha * bn_4, bn_4)
# dp_4 = tf.nn.dropout(lrelu_4, keep_prob)
# Logits
logits = tf.layers.conv2d_transpose(dp_3, out_channel_dim, 4, strides_2, padding_1, kernel_initializer=tf.contrib.layers.xavier_initializer(uniform=True, seed=None, dtype=tf.float32))
# Output
output = tf.tanh(logits)
return output
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_generator(generator, tf)
损失函数（Loss）¶
部署 model_loss 函数训练并计算 GANs 的损失。该函数应返回形如 (discriminator loss, generator loss) 的元组。
使用你已实现的函数：
discriminator(images, reuse=False)generator(z, out_channel_dim, is_train=True)
In [11]:
def model_loss(input_real, input_z, out_channel_dim):
"""
Get the loss for the discriminator and generator
:param input_real: Images from the real dataset
:param input_z: Z input
:param out_channel_dim: The number of channels in the output image
:return: A tuple of (discriminator loss, generator loss)
"""
# TODO: Implement Function
# Define smooth value
smooth = 0.1
# Build up GAN model with discriminator and generator function
# Create generator
GAN_g = generator(input_z, out_channel_dim, is_train = True)
# Create discriminator
GAN_d_real, d_logits_real = discriminator(input_real)
GAN_d_fake, d_logits_fake = discriminator(GAN_g, reuse = True)
# Build up loss calculation
# Calculate the loss of real and fake individually
d_loss_real = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits = d_logits_real,
labels = tf.ones_like(GAN_d_real) * (1-smooth)))
d_loss_fake = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits = d_logits_fake,
labels = tf.zeros_like(GAN_d_fake) * (1-smooth)))
# Calculate the combined loss of generator and discriminator
g_loss = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits = d_logits_fake, labels = tf.ones_like(GAN_d_fake)))
d_loss = d_loss_real + d_loss_fake
return d_loss, g_loss
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_loss(model_loss)
优化（Optimization）¶
部署 model_opt 函数实现对 GANs 的优化。使用 tf.trainable_variables 获取可训练的所有变量。通过变量空间名 discriminator 和 generator 来过滤变量。该函数应返回形如 (discriminator training operation, generator training operation) 的元组。
In [12]:
def model_opt(d_loss, g_loss, learning_rate, beta1):
"""
Get optimization operations
:param d_loss: Discriminator loss Tensor
:param g_loss: Generator loss Tensor
:param learning_rate: Learning Rate Placeholder
:param beta1: The exponential decay rate for the 1st moment in the optimizer
:return: A tuple of (discriminator training operation, generator training operation)
"""
# TODO: Implement Function
# Get all trainable variables
t_vars = tf.trainable_variables()
d_vars = [var for var in t_vars if var.name.startswith('discriminator')]
g_vars = [var for var in t_vars if var.name.startswith('generator')]
update_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS)
g_updates = [opt for opt in update_ops if opt.name.startswith('generator')]
with tf.control_dependencies(g_updates):
d_train_opt = tf.train.AdamOptimizer(learning_rate, beta1=beta1).minimize(d_loss, var_list=d_vars)
g_train_opt = tf.train.AdamOptimizer(learning_rate, beta1=beta1).minimize(g_loss, var_list=g_vars)
return d_train_opt, g_train_opt
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_opt(model_opt, tf)
In [13]:
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import numpy as np
def show_generator_output(sess, n_images, input_z, out_channel_dim, image_mode):
"""
Show example output for the generator
:param sess: TensorFlow session
:param n_images: Number of Images to display
:param input_z: Input Z Tensor
:param out_channel_dim: The number of channels in the output image
:param image_mode: The mode to use for images ("RGB" or "L")
"""
cmap = None if image_mode == 'RGB' else 'gray'
z_dim = input_z.get_shape().as_list()[-1]
example_z = np.random.uniform(-1, 1, size=[n_images, z_dim])
samples = sess.run(
generator(input_z, out_channel_dim, False),
feed_dict={input_z: example_z})
images_grid = helper.images_square_grid(samples, image_mode)
pyplot.imshow(images_grid, cmap=cmap)
pyplot.show()
训练¶
部署 train 函数以建立并训练 GANs 模型。记得使用以下你已完成的函数：
model_inputs(image_width, image_height, image_channels, z_dim)model_loss(input_real, input_z, out_channel_dim)model_opt(d_loss, g_loss, learning_rate, beta1)
使用 show_generator_output 函数显示 generator 在训练过程中的输出。
注意：在每个批次 (batch) 中运行 show_generator_output 函数会显著增加训练时间与该 notebook 的体积。推荐每 100 批次输出一次 generator 的输出。
In [14]:
def train(epoch_count, batch_size, z_dim, learning_rate, beta1, get_batches, data_shape, data_image_mode):
"""
Train the GAN
:param epoch_count: Number of epochs
:param batch_size: Batch Size
:param z_dim: Z dimension
:param learning_rate: Learning Rate
:param beta1: The exponential decay rate for the 1st moment in the optimizer
:param get_batches: Function to get batches
:param data_shape: Shape of the data
:param data_image_mode: The image mode to use for images ("RGB" or "L")
"""
# TODO: Build Model
# Get image parameters
image_batches, image_width, image_height, image_channels = data_shape
# Extract data from data_shape
input_real, input_z, lr = model_inputs(image_width, image_height, image_channels, z_dim)
# Apply loss function
d_loss, g_loss = model_loss(input_real, input_z, image_channels)
# Apply optimization function
d_train_opt, g_train_opt = model_opt(d_loss, g_loss, lr, beta1)
# Parameters initialization
count = 0
show_cycle = 100
loss_cycle = 10
images_num = 25
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
for epoch_i in range(epoch_count):
for batch_images in get_batches(batch_size):
# TODO: Train Model
# Skip the first round and iterate
count += 1
# Range from [-0.5, 0.5] tp [-1, 1]
batch_images = batch_images * 2
# Noise adding for generator
batch_z = np.random.uniform(-1, 1, size = (batch_size, z_dim))
# Optimizing
_ = sess.run(d_train_opt, feed_dict = {input_real: batch_images, input_z: batch_z, lr: learning_rate})
_ = sess.run(g_train_opt, feed_dict = {input_real: batch_images, input_z: batch_z, lr: learning_rate})
# Print generator output and overall training parameters
if count % show_cycle == 0:
show_generator_output(sess, images_num, input_z, image_channels, data_image_mode)
# Print the loss of discriminator and generator
if count % loss_cycle == 0:
d_loss_training = d_loss.eval({input_z: batch_z, input_real: batch_images})
g_loss_training = g_loss.eval({input_z: batch_z, input_real: batch_images})
print("Epoch {}/{}...".format(epoch_i + 1, epochs),
"Discriminator Loss: {:.6f}...".format(d_loss_training),
"Generator Loss: {:.6f}".format(g_loss_training))
MNIST¶
在 MNIST 上测试你的 GANs 模型。经过 2 次迭代，GANs 应该能够生成类似手写数字的图像。确保生成器 (generator) 低于辨别器 (discriminator) 的损失，或接近 0。
In [ ]:
batch_size = 64
z_dim = 64
learning_rate = 0.0002
beta1 = 0.5
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
epochs = 2
mnist_dataset = helper.Dataset('mnist', glob(os.path.join(data_dir, 'mnist/*.jpg')))
with tf.Graph().as_default():
train(epochs, batch_size, z_dim, learning_rate, beta1, mnist_dataset.get_batches,
mnist_dataset.shape, mnist_dataset.image_mode)
CelebA¶
在 CelebA 上运行你的 GANs 模型。在一般的GPU上运行每次迭代大约需要 20 分钟。你可以运行整个迭代，或者当 GANs 开始产生真实人脸图像时停止它。
In [ ]:
batch_size = 32
z_dim = 128
learning_rate = 0.0002
beta1 = 0.5
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
epochs = 1
celeba_dataset = helper.Dataset('celeba', glob(os.path.join(data_dir, 'img_align_celeba/*.jpg')))
with tf.Graph().as_default():
train(epochs, batch_size, z_dim, learning_rate, beta1, celeba_dataset.get_batches,
celeba_dataset.shape, celeba_dataset.image_mode)
提交项目¶
提交本项目前，确保运行所有 cells 后保存该文件。
保存该文件为 "dlnd_face_generation.ipynb"， 并另存为 HTML 格式 "File" -> "Download as"。提交项目时请附带 "helper.py" 和 "problem_unittests.py" 文件。